Deprecate C++03, C++11, MSVC < 2017, GCC < 5.0

Build infrastructure and static configuration fixes:

- Bump CMAKE_CXX_STANDARD to 14
- Add `-Werror all-warnings` to NVCC to promote warnings to errors
- Add `-Xcudafe --display_error_number` to get useful diagnositics from
    cudafe.
- Clean up cub include dir spec in CMake.
- Move THRUST_DEPRECATED logic out of compiler.h and into new header.
- Fix CPP dialect detection on newer MSVC.
- Remove raw `__cplusplus` checks.
- Use `_Pragma`/`__pragma` instead of `#pragma` in macro.
- Remove THRUST_BEGIN/END_NS macros.
  - These were used inconsistently, rendering them non-functional. Removing
    to prevent people from trying to use them.

Workarounds for msvc:

- MSVC isn't a fan of `decltype(...)::some_member` syntax.
  - WAR by aliasing the `decltype(...)` and doing `NewAlias::some_member`
- Missing `template` keyword when rebinding pointer in `async/reduce.h`
- Silence warning C4494 `declspec(allocator) used on non-pointer/ref type`
  - Bug in MSVC STL: microsoft/STL#696
- Disable async sort test on MSVC
  - Triage. Looks like a bug in cudafe? See NVIDIA#1098.
- Add pointer<T>::pointer_to(reference)
  - Required for C++11, hard compile error on MSVC.
- Bring a definition of `atanh` into scope for complex number impl
- Fix floating point literals be declared as floats instead of doubles
- Replace `std::remove_reference<T>::type&` with
  `std::add_lvalue_reference`.
  - Same behavior, and MSVC chokes on the other syntax when followed by
    `__host__`.
- Remove constexpr markup from defaulted functions.
  - These are constexpr by default when possible, and the compilers were
    complaining about the markup in places.
- Use `thrust::detail::integer_traits` instead of `std::numeric_limits`
  in device code.
- Avoid aligning beyond platform limits in alignment.cu.
- Pass /bigobj to MSVC so it can handle the async tests
- Work around MSVC compiler bug by replacing SFINAE with static dispatch

Bug 2865172
Bug 2880936

CMakeLists.txt

@@ -64,11 +64,13 @@ endif ()
 add_definitions(-DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_${THRUST_DEVICE_SYSTEM})
 
 # Please note this also sets the default for the CUDA C++ version; see the comment below.
-set(CMAKE_CXX_STANDARD 11 CACHE STRING "The C++ version to be used.")
+set(CMAKE_CXX_STANDARD 14 CACHE STRING "The C++ version to be used.")
 set(CMAKE_CXX_EXTENSIONS OFF)
 
 message("-- C++ Standard version: ${CMAKE_CXX_STANDARD}")
 
+set(CUB_INCLUDE_DIR "${PROJECT_SOURCE_DIR}/dependencies/cub")
+
 if ("CUDA" STREQUAL "${THRUST_DEVICE_SYSTEM}")
   if (NOT "${CMAKE_CUDA_HOST_COMPILER}" STREQUAL "")
     unset(CMAKE_CUDA_HOST_COMPILER CACHE)
@@ -177,6 +179,19 @@ if ("MSVC" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
   # Disable warning about applying unary operator- to unsigned type.
   append_option_if_available("/wd4146" THRUST_CXX_WARNINGS)
 
+  # MSVC STL assumes that `allocator_traits`'s allocator will use raw pointers,
+  # and the `__DECLSPEC_ALLOCATOR` macro causes issues with thrust's universal
+  # allocators:
+  #   warning C4494: 'std::allocator_traits<_Alloc>::allocate' :
+  #      Ignoring __declspec(allocator) because the function return type is not
+  #      a pointer or reference
+  # See https://github.com/microsoft/STL/issues/696
+  append_option_if_available("/wd4494" THRUST_CXX_WARNINGS)
+
+  # Some of the async tests require /bigobj to fit all their sections into the
+  # object files:
+  append_option_if_available("/bigobj" THRUST_CXX_WARNINGS)
+
   set(THRUST_TREAT_FILE_AS_CXX "/TP")
 else ()
   append_option_if_available("-Werror" THRUST_CXX_WARNINGS)
@@ -243,6 +258,8 @@ if ("CUDA" STREQUAL "${THRUST_DEVICE_SYSTEM}")
   foreach (CXX_OPTION IN LISTS THRUST_CXX_WARNINGS)
     set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler=${CXX_OPTION}")
   endforeach ()
+  set(CMAKE_CUDA_FLAGS
+    "${CMAKE_CUDA_FLAGS} -Werror all-warnings -Xcudafe --display_error_number")
 endif ()
 
 # For every public header, build a translation unit containing `#include <header>`
@@ -355,7 +372,7 @@ endforeach ()
 add_library(header-test OBJECT ${THRUST_HEADER_TEST_SOURCES})
 target_include_directories(
   header-test
-  PUBLIC ${PROJECT_SOURCE_DIR}
+  PUBLIC ${PROJECT_SOURCE_DIR} ${CUB_INCLUDE_DIR}
 )
 
 include(CTest)
@@ -383,7 +400,7 @@ endif ()
 add_library(thrust_testframework STATIC ${THRUST_TESTFRAMEWORK_FILES})
 target_include_directories(
   thrust_testframework
-  PUBLIC ${PROJECT_SOURCE_DIR} ${PROJECT_SOURCE_DIR}/dependencies/cub
+  PUBLIC ${PROJECT_SOURCE_DIR} ${CUB_INCLUDE_DIR}
   PRIVATE ${PROJECT_SOURCE_DIR}/testing
 )
 
@@ -491,7 +508,7 @@ foreach (THRUST_TEST_SOURCE IN LISTS THRUST_TESTS)
 
   target_include_directories(
     ${THRUST_TEST}
-    PUBLIC ${PROJECT_SOURCE_DIR} ${PROJECT_SOURCE_DIR}/dependencies/cub
+    PUBLIC ${PROJECT_SOURCE_DIR} ${CUB_INCLUDE_DIR}
     PRIVATE ${PROJECT_SOURCE_DIR}/testing
   )
 
@@ -518,7 +535,7 @@ foreach (THRUST_TEST_SOURCE IN LISTS THRUST_TESTS)
 
     target_include_directories(
       ${THRUST_TEST_RDC}
-      PUBLIC ${PROJECT_SOURCE_DIR} ${PROJECT_SOURCE_DIR}/dependencies/cub
+      PUBLIC ${PROJECT_SOURCE_DIR} ${CUB_INCLUDE_DIR}
       PRIVATE ${PROJECT_SOURCE_DIR}/testing
     )
 
@@ -617,7 +634,7 @@ foreach (THRUST_EXAMPLE_SOURCE IN LISTS THRUST_EXAMPLES)
 
   target_include_directories(
     ${THRUST_EXAMPLE}
-    PUBLIC ${PROJECT_SOURCE_DIR} ${PROJECT_SOURCE_DIR}/dependencies/cub
+    PUBLIC ${PROJECT_SOURCE_DIR} ${CUB_INCLUDE_DIR}
     PRIVATE ${PROJECT_SOURCE_DIR}/examples
   )
 
@@ -640,7 +657,7 @@ foreach (THRUST_EXAMPLE_SOURCE IN LISTS THRUST_EXAMPLES)
 
     target_include_directories(
       ${THRUST_EXAMPLE_RDC}
-      PUBLIC ${PROJECT_SOURCE_DIR} ${PROJECT_SOURCE_DIR}/dependencies/cub
+      PUBLIC ${PROJECT_SOURCE_DIR} ${CUB_INCLUDE_DIR}
       PRIVATE ${PROJECT_SOURCE_DIR}/examples
     )
 

doc/thrust.dox

@@ -2063,8 +2063,6 @@ PREDEFINED             = THRUST_NOEXCEPT=noexcept \
                          "THRUST_MR_DEFAULT_ALIGNMENT=alignof(max_align_t)" \
                          "THRUST_FINAL=final" \
                          "THRUST_OVERRIDE=" \
-                         "THRUST_BEGIN_NS=namespace thrust {" \
-                         "THRUST_END_NS=}" \
                          "cuda_cub=system::cuda"
 
 # If the MACRO_EXPANSION and EXPAND_ONLY_PREDEF tags are set to YES then this

examples/cuda/async_reduce.cu

@@ -1,9 +1,10 @@
+#include <thrust/detail/config.h>
 #include <thrust/device_vector.h>
 #include <thrust/reduce.h>
 #include <thrust/system/cuda/execution_policy.h>
 #include <cassert>
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 #include <future>
 #endif
 
@@ -52,7 +53,7 @@ int main()
   // reset the result
   result[0] = 0;
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
   // method 2: use std::async to create asynchrony
 
   // copy all the algorithm parameters

examples/cuda/global_device_vector.cu

@@ -1,3 +1,4 @@
+#include <thrust/detail/config.h>
 #include <thrust/device_vector.h>
 
 // If you create a global `thrust::device_vector` with the default allocator,
@@ -20,7 +21,7 @@ typedef thrust::system::cuda::detail::cuda_memory_resource<
   thrust::cuda::pointer<void>
 > device_ignore_shutdown_memory_resource;
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
   template <typename T>
   using device_ignore_shutdown_allocator = 
     thrust::mr::stateless_resource_allocator<

internal/benchmark/bench.cu

@@ -4,6 +4,7 @@
 #include <thrust/sort.h>
 #include <thrust/reduce.h>
 #include <thrust/scan.h>
+#include <thrust/detail/config.h>
 
 #if THRUST_CPP_DIALECT >= 2011
 #include <thrust/shuffle.h>
@@ -47,7 +48,7 @@
 
 // We don't use THRUST_NOEXCEPT because it's new, and we want this benchmark to
 // be backwards-compatible to older versions of Thrust.
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
   #define NOEXCEPT noexcept
 #else
   #define NOEXCEPT throw()

internal/build/common_warnings.mk

@@ -120,5 +120,16 @@ else ifeq ($(OS),win32)
 
   # Disable warning about applying unary - to unsigned type.
   CUDACC_FLAGS += -Xcompiler "/wd4146"
+
+  # Warning about declspec(allocator) on inappropriate function types
+  CUDACC_FLAGS += -Xcompiler "/wd4494"
+
+  # Allow tests to have lots and lots of sections in each translation unit:
+  CUDACC_FLAGS += -Xcompiler "/bigobj"
 endif
 
+# Promote all NVCC warnings into errors
+CUDACC_FLAGS += -Werror all-warnings
+
+# Print warning numbers with cudafe diagnostics
+CUDACC_FLAGS += -Xcudafe --display_error_number

testing/alignment.cu

@@ -210,14 +210,31 @@ void test_aligned_type()
 DECLARE_UNITTEST(test_aligned_type);
 
 template <std::size_t Len, std::size_t Align>
-void test_aligned_storage_instantiation()
+void test_aligned_storage_instantiation(thrust::detail::true_type /* Align is valid */)
 {
     typedef typename thrust::detail::aligned_storage<Len, Align>::type type;
     ASSERT_GEQUAL(sizeof(type), Len);
     ASSERT_EQUAL(THRUST_ALIGNOF(type), Align);
     ASSERT_EQUAL(thrust::detail::alignment_of<type>::value, Align);
 }
 
+template <std::size_t Len, std::size_t Align>
+void test_aligned_storage_instantiation(thrust::detail::false_type /* Align is invalid */)
+{
+  std::cerr << "Skipping test_aligned_storage_instantiation<Len=" << Len
+            << ", Align=" << Align << "> (MaxAlign="
+            << THRUST_ALIGNOF(thrust::detail::max_align_t) << ").\n";
+}
+
+template <std::size_t Len, std::size_t Align>
+void test_aligned_storage_instantiation()
+{
+  typedef thrust::detail::integral_constant<
+      bool, Align <= THRUST_ALIGNOF(thrust::detail::max_align_t)>
+      ValidAlign;
+  test_aligned_storage_instantiation<Len, Align>(ValidAlign());
+}
+
 template <std::size_t Len>
 void test_aligned_storage_size()
 {

testing/allocator.cu

@@ -1,4 +1,5 @@
 #include <unittest/unittest.h>
+#include <thrust/detail/config.h>
 #include <thrust/device_malloc_allocator.h>
 #include <thrust/system/cpp/vector.h>
 #include <memory>
@@ -202,7 +203,7 @@ void TestAllocatorTraitsRebind()
 }
 DECLARE_UNITTEST(TestAllocatorTraitsRebind);
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 void TestAllocatorTraitsRebindCpp11()
 {
   ASSERT_EQUAL(
@@ -250,5 +251,5 @@ void TestAllocatorTraitsRebindCpp11()
   );
 }
 DECLARE_UNITTEST(TestAllocatorTraitsRebindCpp11);
-#endif
+#endif // C++11
 

testing/async_sort.cu

@@ -1,6 +1,8 @@
 #include <thrust/detail/config.h>
 
-#if THRUST_CPP_DIALECT >= 2011 && !defined(THRUST_LEGACY_GCC)
+// Disabled on MSVC for GH issue #1098
+#if THRUST_CPP_DIALECT >= 2011 && !defined(THRUST_LEGACY_GCC) && \
+  THRUST_HOST_COMPILER != THRUST_HOST_COMPILER_MSVC
 
 #include <unittest/unittest.h>
 

testing/complex.cu

@@ -1,6 +1,8 @@
 #include <unittest/unittest.h>
 
 #include <thrust/complex.h>
+#include <thrust/detail/config.h>
+
 #include <complex>
 #include <iostream>
 #include <sstream>
@@ -273,7 +275,7 @@ struct TestComplexTrigonometricFunctions
     ASSERT_ALMOST_EQUAL(sinh(a),sinh(c));
     ASSERT_ALMOST_EQUAL(tanh(a),tanh(c));
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 
     ASSERT_ALMOST_EQUAL(acos(a),acos(c));
     ASSERT_ALMOST_EQUAL(asin(a),asin(c));

testing/dependencies_aware_policies.cu

@@ -1,5 +1,6 @@
 #include <unittest/unittest.h>
 
+#include <thrust/detail/config.h>
 #include <thrust/detail/seq.h>
 #include <thrust/system/cpp/detail/par.h>
 #include <thrust/system/omp/detail/par.h>
@@ -9,7 +10,7 @@
 #  include <thrust/system/cuda/detail/par.h>
 #endif
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 
 template<typename T>
 struct test_allocator_t
@@ -178,11 +179,11 @@ SimpleUnitTest<
     >
 > TestDependencyAttachmentInstance;
 
-#else
+#else // C++11
 
 void TestDummy()
 {
 }
 DECLARE_UNITTEST(TestDummy);
 
-#endif
+#endif // C++11

testing/mr_disjoint_pool.cu

@@ -1,8 +1,10 @@
 #include <unittest/unittest.h>
+
+#include <thrust/detail/config.h>
 #include <thrust/mr/disjoint_pool.h>
 #include <thrust/mr/new.h>
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 #include <thrust/mr/disjoint_sync_pool.h>
 #endif
 
@@ -177,7 +179,7 @@ void TestDisjointUnsynchronizedPool()
 }
 DECLARE_UNITTEST(TestDisjointUnsynchronizedPool);
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 void TestDisjointSynchronizedPool()
 {
     TestDisjointPool<thrust::mr::disjoint_synchronized_pool_resource>();
@@ -260,7 +262,7 @@ void TestDisjointUnsynchronizedPoolCachingOversized()
 }
 DECLARE_UNITTEST(TestDisjointUnsynchronizedPoolCachingOversized);
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 void TestDisjointSynchronizedPoolCachingOversized()
 {
     TestDisjointPoolCachingOversized<thrust::mr::disjoint_synchronized_pool_resource>();
@@ -285,7 +287,7 @@ void TestUnsynchronizedDisjointGlobalPool()
 }
 DECLARE_UNITTEST(TestUnsynchronizedDisjointGlobalPool);
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 void TestSynchronizedDisjointGlobalPool()
 {
     TestDisjointGlobalPool<thrust::mr::disjoint_synchronized_pool_resource>();

testing/mr_pool.cu

@@ -1,8 +1,10 @@
 #include <unittest/unittest.h>
+
+#include <thrust/detail/config.h>
 #include <thrust/mr/pool.h>
 #include <thrust/mr/new.h>
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 #include <thrust/mr/sync_pool.h>
 #endif
 
@@ -241,7 +243,7 @@ void TestUnsynchronizedPool()
 }
 DECLARE_UNITTEST(TestUnsynchronizedPool);
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 void TestSynchronizedPool()
 {
     TestPool<thrust::mr::synchronized_pool_resource>();
@@ -324,7 +326,7 @@ void TestUnsynchronizedPoolCachingOversized()
 }
 DECLARE_UNITTEST(TestUnsynchronizedPoolCachingOversized);
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 void TestSynchronizedPoolCachingOversized()
 {
     TestPoolCachingOversized<thrust::mr::synchronized_pool_resource>();
@@ -348,7 +350,7 @@ void TestUnsynchronizedGlobalPool()
 }
 DECLARE_UNITTEST(TestUnsynchronizedGlobalPool);
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
 void TestSynchronizedGlobalPool()
 {
     TestGlobalPool<thrust::mr::synchronized_pool_resource>();

testing/vector.cu

@@ -1,6 +1,9 @@
 #include <unittest/unittest.h>
+
+#include <thrust/detail/config.h>
 #include <thrust/sequence.h>
 #include <thrust/device_malloc_allocator.h>
+
 #include <vector>
 #include <list>
 #include <limits>
@@ -742,7 +745,7 @@ void TestVectorReversed(void)
 }
 DECLARE_VECTOR_UNITTEST(TestVectorReversed);
 
-#if __cplusplus >= 201103L
+#if THRUST_CPP_DIALECT >= 2011
   template <class Vector>
   void TestVectorMove(void)
   {